Vaccine data, Cases data from the hospital, and the waste water
signal data has been loaded, cleaned, and then merged into one final
dataframe, final_data.
The variables have been log transformed.
The response variable observed_census_ICU_p_acute_care
has been renamed as y and the date has been renamed as ds to fit the
prophet model.
The dataset is divided into train and test set. The test set consist of last 3 days of data.
Adding other variable as regressors to the model.
Fitting the Prophet Model.
Forecasting 3 days into future
Checking last 6 days of the forecast data
#> ds yhat yhat_lower yhat_upper
#> 407 2022-01-27 4.409765 3.477338 5.338548
#> 408 2022-01-28 4.427693 3.494753 5.312179
#> 409 2022-01-29 4.419779 3.556612 5.322547
#> 410 2022-01-30 4.550004 3.688042 5.473451
#> 411 2022-01-31 4.369424 3.461819 5.264422
#> 412 2022-02-01 4.461659 3.572634 5.360985
The plot of actual data and predicted data from Prophet forecast. The blue line is predicted data whereas the black dots are actual data.
Root Mean Squared Error on training data:
#> [1] 16.89343
MAPE on train set:
#> [1] 0.3981676
Standard deviation of the actual data
#> [1] 35.92902
Plots comparing actual data and predicted data
RMSE on test set:
#> [1] 9.341877
MAPE on test set
#> [1] 0.1049049
Plots comparing actual data and predicted data
The error metrics are low when model is regressed against only those who received 1st dose in all the different age groups.
The model trend shows a strong linear decrease in trend of hospitalizations in 2021 until July/August 2021 and from August to December there is a moderately strong linear increase in hospitalizations. The extra regressor plot shows the additive effect of regressors and it shows that 1st dose of vaccination in all different age groups shows a linear increase until April/May 2021 and then shows a slight decrease in July 2021 and then remains low and steady until January 2022. Weekly trend shows there are more hospitalizations on Tuesday and Thursday.